Characterization of the substrate binding site in rat liver 3alpha-hydroxysteroid/dihydrodiol dehydrogenase. The roles of tryptophans in ligand binding and protein fluorescence.

Rat liver 3a-hydroxysteroid dehydrogenase (3a-HSD), a member of the aldoketoreductase superfamily, inactivates circulating steroid hormones using NAD(P)H as cofactor. Despite determination of the 3a-HSDzNADP binary complex structure, the functional elements that dictate the binding of steroids remain unclear (Bennett, M.J., Schlegel, B.P., Jez, J.M., Penning, T.M., and Lewis, M. (1996) Biochemistry 35, 10702–10711). Two tryptophans (Trp and Trp) near the active site may have roles in substrate binding, and their fluorescence may be quenched upon binding of NADPH. Trp is located within an apolar cleft, while Trp is found on an opposing loop near the active site. A third tryptophan, Trp, is on the periphery of the structure. To investigate the roles of these tryptophans in protein fluorescence and ligand binding, we generated three mutant enzymes (W86Y, W148Y, and W227Y) by site-directed mutagenesis. Spectroscopic measurements on these proteins showed that Trp contributed the most to the enzyme fluorescence spectra, with Trp adding the least. Trp was identified as the tryptophan quenched by bound NADPH through an energy transfer mechanism. The W86Y mutant altered binding of cofactor (a 3-fold increase in Kd for NADPH) and steroid (a 7-fold increase in Kd for testosterone). This mutation also dramatically decreased the catalytic efficiency observed with one-, two-, and three-ring substrates and decreased the binding affinity for nonsteroidal anti-inflammatory drugs but had little effect on the binding of aldose reductase inhibitors. Interestingly, mutation of Trp significantly impaired steroid binding (a 22-fold increase in Kd for testosterone), but did not alter binding of cofactor, smaller substrates, or inhibitors. Kinetically, the W148Y mutant was similar to wild-type enzyme. Our results demonstrate that Trp and the apolar cleft is part of the substrate binding pocket. In addition, we propose a role for Trp and its associated loop in binding steroids, but not small substrates or inhibitors, most likely through interaction with the Cand D-rings of the steroid. This work provides the first evidence that tryptophans on opposite sides of the apolar cleft are part of the steroid binding pocket and suggests how the enzyme may discriminate between nonsteroidal anti-inflammatory drugs and aldose reductase inhibitors like zopolrestat. A model of how androstanedione binds in the apolar cleft is developed. These data provide further evidence that loop structures in members of the aldoketoreductase superfamily are critical determinants of ligand binding.

Rat liver 3␣-hydroxysteroid dehydrogenase (3␣-HSD, EC 1.1.1.213) 1 is a representative hydroxysteroid dehydrogenase (HSD) with the principal physiological role of inactivating circulating androgens, progestins, and glucocorticoids. It also functions as a dihydrodiol dehydrogenase by oxidizing polycyclic aromatic hydrocarbon trans-dihydrodiols (proximate carcinogens) into ortho-quinones with the production of reactive oxygen species and semiquinone radicals that may contribute to chemical carcinogenesis (1)(2)(3). It is very similar (Ͼ69% amino acid sequence identity) to its human homologues, including human type I and type II 3␣-HSDs (4 -6). In endocrine target tissues, like the prostate, 3␣-HSD converts 5␣-dihydrotestosterone (a potent androgen) into 3␣-androstanediol (a weak androgen) (7) and may act as a molecular switch by regulating occupancy of the androgen receptor. As the most thoroughly characterized mammalian 3␣-HSD, rat liver 3␣-HSD serves as an excellent model for investigating the structure and function of these enzymes.
cDNA cloning indicates that the mammalian 3␣-HSDs are members of the aldoketoreductase (AKR) superfamily and are 45-60% identical in amino acid sequence to other AKR proteins, including aldose reductase, aldehyde reductase, 17␤-HSD, and 20␣-HSD (8 -12). This similarity raises the issue of how closely related enzymes recognize different substrates (aldo-keto sugars versus steroid hormones) and is an important concern in developing therapeutic agents against specific protein targets. For example, retinopathic, neuropathic, and nephropathic diabetic complications have been associated with the conversion of glucose to sorbitol catalyzed by aldose reductase (13). As such, aldose reductase inhibitors may be useful therapies for these complications of diabetes, but the specificity of existing compounds has been lacking (14). Similarly, the potential role of 3␣-HSDs in the regulation of hormone levels in endocrine target tissues and in carcinogen activation make these enzymes candidate drug targets. Although nonsteroidal anti-inflammatory drugs (NSAIDs) inhibit rat liver 3␣-HSD, these compounds are not specific for this protein (15). An understanding of how these enzymes achieve substrate specificity would provide insight into designing effective and specific AKR inhibitors.
The three-dimensional structures of the rat liver 3␣-HSD apoenzyme and the E⅐NADP ϩ binary complex have been solved at 3.0-and 2.7-Å resolutions, respectively (16 -17). This protein adopts an (␣/␤) 8 -barrel fold and contains three tryptophans (Trp 86 , Trp 148 , and Trp 227 ). Although the location of the cofactor in the binary complex and site-directed mutagenesis studies have identified the residues involved in catalysis (18), the functional components of the substrate binding site remain unknown. The active site is at the base of an apolar cleft. Projecting toward the center of the barrel, this cleft is about 11 Å deep, is large enough to accommodate steroid ligands, and could be the substrate binding site in rat liver 3␣-HSD. Trp 86 and other apolar residues (Leu 54 , Phe 128 , and Phe 129 ) form one side of this cleft and would provide an ideal environment for binding steroid hormones. A flexible loop containing Trp 227 may form part of the opposite side of the substrate binding pocket. However, this loop was disordered in the electron density of the apoenzyme structure, and in the binary complex structure Trp 227 made a crystal contact with the second molecule in the asymmetric unit. Neither structure clearly indicates the location of Trp 227 or the flexible loop in relation to the putative steroid binding site. Finally, binding of NADPH quenches the fluorescence emission of 3␣-HSD, suggesting that either Trp 86 or Trp 227 may be near bound cofactor (18,19). The third tryptophan, Trp 148 , should have no function in ligand binding based on its location away from the active site.
In this work, we used site-directed mutagenesis to investigate the role of tryptophans in protein fluorescence and to delineate the structural components of the steroid and inhibitor binding site by constructing the following mutants of rat liver 3␣-HSD: W86Y, W148Y, and W227Y. Our results indicate that the contribution of each tryptophan to the overall fluorescence of 3␣-HSD corresponds to the local environment of each residue, with Trp 148 (the most buried tryptophan) dominating the spectrum and Trp 227 (the most solvent-exposed tryptophan) providing the smallest contribution. We demonstrate that an energy transfer mechanism quenches Trp 86 upon binding of NADPH. Our work provides the first functional evidence that Trp 86 is important in binding steroids and NSAIDs and that the apolar cleft in which it resides is part of the substrate/ inhibitor binding pocket. The mutagenesis data also suggest that Trp 227 plays an important role in binding steroid hormones but has no role in accommodating one-, two-, or threering substrates or NSAIDs. Also, mutations of Trp 86 and Trp 227 had little effect on the binding of aldose reductase inhibitors. A model for binding androstanedione and how the enzyme may discriminate between NSAIDs and aldose reductase inhibitors is proposed. These data provide another example where loop structures in (␣/␤) 8 -barrel proteins are determinants in ligand binding.

EXPERIMENTAL PROCEDURES
Materials-The DNA Synthesis Service in the Department of Chemistry at the University of Pennsylvania synthesized the primers used for polymerase chain reaction-based site-directed mutagenesis. The DE-52 cellulose was from Baxter; and the Blue Sepharose was purchased from Pharmacia. Goat anti-rabbit IgG-horseradish peroxidase conjugate and 4-chloro-1-naphthol were from Bio-Rad. Smithgall and Penning (20) previously described the preparation of polyclonal rabbit anti-rat 3␣-HSD antiserum. NAD ϩ , NADH, and NADPH were from Boehringer-Mannheim. Androsterone, androstanedione, and testosterone were obtained from Steraloids. Radiolabeled [ 3 H]testosterone (92.5 Ci/mmol) was purchased from NEN DuPont. Zopolrestat and ponalrestat were provided courtesy of Dr. Florante Quiocho. All other compounds were ACS grade or better and obtained from Sigma or Aldrich.
Mutagenesis, Expression, and Purification of Recombinant Wild-type and Mutant 3␣-HSDs-The pKK-3␣-HSD expression vector and details of the polymerase chain reaction-based site-directed mutagenesis protocols were previously described (18). Site-directed mutagenesis to produce the W86Y, W148Y, and W227Y mutant enzymes used the following oligonucleotide primers, respectively: 5Ј-dCTTCAAAGCTTT-ATAGCACTTTCCA-3Ј; 5Ј-dACTTGTGACACATATGAGGCCATGG-3Ј; and 5Ј-dGAGACAAAACATATGTGGATCAGAA-3Ј. Dideoxysequencing ensured fidelity of the mutant constructs. The mutant expression vectors transformed competent Escherichia coli DH5␣ cells, and the overexpressed proteins were purified. Four-liter cultures of E. coli cells containing pKK-3␣-HSD (either wild-type or mutant) grown in LB media with 100 g/ml ampicillin for 30 h at 37°C were pelleted, resuspended, and lysed by sonication. Pooled sonicates were dialyzed overnight in 10 mM Tris-HCl buffer (pH 8.6) containing 1 mM EDTA, 1 mM ␤-mercaptoethanol (BME), and 20% glycerol. The dialyzed fraction was loaded onto a DE-52 cellulose anion exchange column equilibrated with the same buffer and eluted using a 0 -250 mM NaCl salt gradient. SDS-PAGE and assays for androsterone oxidation located the peak fractions. These fractions were pooled and dialyzed overnight against 10 mM potassium phosphate buffer (pH 7.0) containing 1 mM EDTA, 1 mM BME, and 20% glycerol and loaded onto a Blue Sepharose column equilibrated in dialysis buffer. The column was washed in dialysis buffer, and the enzyme was eluted with 1.5 M KCl. Eluted fractions were dialyzed overnight against 20 mM potassium phosphate buffer (pH 7.0) containing 1 mM EDTA, 1 mM BME, and 30% glycerol. SDS-PAGE was used to analyze protein purity (21), and protein concentrations were determined by the Lowry method (22). Immunoblots confirmed the identity of the expressed protein using a 1:1000 dilution of the polyclonal rabbit anti-rat 3␣-HSD sera 71535. Visualization of proteins used rabbit anti-goat IgG-horseradish peroxidase conjugate with 4chloro-1-naphthol as chromogenic substrate. Samples of the homogeneous enzymes were stored as aliquots at Ϫ70°C.
Spectroscopic Measurements-Measurement of the fluorescence excitation and emission spectra of recombinant wild type and the tryptophan mutants at 25°C were conducted on a Perkin-Elmer LS5 spectrophotometer equipped with an ATT 6300 personal computer. The FLUOR program (Softways, Inc.) digitized the spectra. Scans of the fluorescence excitation and emission spectra were performed in 1-ml systems containing 20 g of protein and 10 mM potassium phosphate (pH 7.0) buffer. Determination of the energy transfer emission spectra of the apoenzymes, the E⅐NADPH binary complexes, and the E⅐NADPH⅐testosterone ternary complexes of wild-type 3␣-HSD and each tryptophan mutant used the same system as above with 4% acetonitrile as co-solvent. The binary complex spectra were measured in the presence of 1.5 M NADPH, while the ternary complex spectra were scanned with 50 M testosterone added.
Steady-state Enzyme Kinetics-Initial velocities were measured on either a Gilford 260 spectrophotometer or a Beckman DU-640 spectrophotometer by observing the rate of change in absorbance of pyridine nucleotide at 340 nm (⑀ ϭ 6270 M Ϫ1 cm Ϫ1 ) in 1-ml systems at 25°C using a 1-cm path length. Throughout the purifications, enzyme activity was monitored using a standard assay system containing 100 mM potassium phosphate buffer (pH 7.0), 2.3 mM NAD ϩ , and 75 M androsterone with 4% acetonitrile as co-solvent. Measurements of the K m and k cat values for androsterone oxidation were made at 2.3 mM NAD ϩ with varied steroid concentration (11.5-73.0 M). K m and k cat values for NAD ϩ were determined at 75 M androsterone by varying the NAD ϩ concentration (0.09 -3.3 mM). Kinetic constants for androstanedione reduction were measured at 180 M NADH with varied steroid concentration (2.4 -28.4 M). Determination of the kinetic constants for NADH oxidation used 30 M androstanedione and were made by varying the NADH concentration (4.3-177 M). These measurements all used the standard assay system. Determination of kinetic constants with 4-nitrobenzaldehyde (0.02-2.0 mM) and 2-decalone (0.2 -5 mM) as substrates used the same assay system at pH 6.0 with 200 M NADH. Assays with 9,10-phenanthrenequinone (1-30 M) used 200 M NADPH as cofactor in the pH 6.0 buffer. All reactions were initiated by the addition of enzyme and were corrected for nonenzymatic rates. Calculation of all k cat and K m values used the ENZFITTER nonlinear regression analysis program (23) to fit untransformed data with a hyperbolic function, as originally described by Wilkinson (24), yielding estimates of the kinetic constants and the associated standard error.
Inhibition Studies-Initial velocities were measured using the standard androsterone assay system as described above with various concen-trations of inhibitors in 20 l of methanol, ethanol, or acetonitrile. Control velocities were determined in the presence of appropriate quantities of organic solvents. Calculations of IC 50 values were from median effect plots (25).
Determination of Binding Constants for NADPH by Fluorescence Titration-Measurements of the binding constants for NADPH to recombinant wild type and the W86Y, W148Y, and W227Y mutants were made by measuring protein fluorescence on a Perkin-Elmer model 650-10M flurometer following the incremental addition of NADPH (0 -2.5 M). Each 1-ml sample contained 20 g of protein in 10 mM potassium phosphate (pH 7.0) buffer at 20°C in a 0.4 ϫ 1.0-cm quartz cuvette. The total volume change from the addition of NADPH was less than 5%, and the necessary corrections for changes in cofactor concentrations were made. Excitation of samples was at 280 nm with fluorescence emission scanned from 300 to 500 nm at 120 nm/min with excitation and emission band pass each set at 5 nm. Untransformed fluorescence data were plotted as the percentage change in fluorescence at emission max (%⌬F) versus NADPH concentration. Fitting of these data to a saturation absorption isotherm by ENZFITTER provided an estimate of the K d and the associated standard errors (23). Transformation of these data used the Lineweaver-Burk equation to generate a linear plot of 1/%⌬F versus 1/[NADPH].
Measurement of Binding Constants for Testosterone to the E⅐NADH Complex-Binding of testosterone to the E⅐NADH binary complex was measured by equilibrium dialysis. Aliquots (100 l) of 10 M 3␣-HSD (either wild-type or mutant) were placed in dialysis tubing with a saturating concentration of NADH (2.5 mM) and no steroid in 10 mM potassium phosphate (pH 7.0) buffer with 1 mM EDTA, 1 mM BME, and 4% acetonitrile as co-solvent. The 1-ml dialysate contained 2.5 mM NADH and increasing concentrations of [ 3 H]testosterone (0.5-61 M; 75,000 cpm/nmol) in an identical buffer. At equilibrium (16 h at 20°C), scintillation counting measured the amount of radiolabeled steroid ligand in both the dialysate and the bag. Since steroid binding does not occur in the absence of NADH, corrections for nonspecific binding were not necessary. Determination of the K d values for the binding of [ 3 H]testosterone to wild-type and mutant 3␣-HSDs used Scatchard analysis with the ENZFITTER program (23).
Molecular Modeling-Molecular modeling on a Silicon Graphics workstation utilized the program QUANTA (Molecular Simulations, Inc.). Bennett et al. (17) reported the structure of the rat liver 3␣-HSD⅐NADP ϩ binary complex; and the coordinates of androstanedione were obtained from the Cambridge Structural Database (26). Androstanedione was manually docked into the apolar cleft of the binary complex structure using the following guidelines: 1) the position of the C3 ketone was placed within hydrogen bonding distance of Tyr 55 , His 117 , and NADP ϩ ; 2) the ␣-face of the steroid was oriented toward the active site to retain the 4-pro-R-stereospecificity of hydride transfer. The loop containing Trp 227 was modeled as in the apoenzyme structure (16), since this loop forms a crystal contact in the binary complex. The model was then energy-minimized using a conjugate gradient algorithm for 100 cycles with CHARMm (Molecular Simulations Inc.).

Expression and Purification of Recombinant Wild-type and
Mutant 3␣-HSDs-Wild-type 3␣-HSD and the W86Y, W148Y, and W227Y mutants were overexpressed in E. coli DH5␣ cells and purified (Table I). The specific activities of the purified proteins for androsterone turnover were 1.50, 0.059, 0.674, and 0.080 mol/min/mg for wild-type enzyme and the W86Y, W148Y, and W227Y mutants, respectively. The wild-type and mutant forms of the 3␣-HSD appeared as single homogeneous bands of the same molecular weight on SDS-PAGE gels (Fig.  1A). Each of the mutants was also immunoreactive with rabbit anti-rat 3␣-HSD antiserum, as demonstrated by Western blot analysis (Fig. 1B).  Fluorescence Excitation and Emission Spectra of Wild-type and Mutant 3␣-HSDs-The 3␣-HSD mutants allowed the contribution from each tryptophan to the overall protein fluorescence spectrum to be assessed. Although the fluorescence signals of the tryptophans were not strictly additive, Trp 148 dominated the spectra, Trp 86 provided the next largest share, and Trp 227 contributed the least.
Removal of each tryptophan also shifted the emission max relative to wild-type enzyme and provided an indication of the local environments of these residues (Fig. 2). Wild-type 3␣-HSD had an emission max ϭ 336 nm. The W86Y and W227Y mutants were blue-shifted with emission max of 332 and 331 nm, respectively, and suggested a partly solvent-accessible environment for these tryptophans (27). The W227Y mutant also displayed a noticeable loss of emission in the higher wavelengths, indicating that this residue is highly solvent-exposed.  Finally, mutation of Trp 148 had the opposite effect; it redshifted the emission max to 343 nm, consistent with this residue being buried away from solvent. These mutations showed that the contribution of each tryptophan to the fluorescence spectra corresponds to their solvent exposure in the threedimensional structure; the more solvent-inaccessible the tryptophan, the greater contribution to the observed spectra. Trp 148 is solvent-inaccessible on the interior of an ␣-helix facing the ␤-barrel, and Trp 86 , in the apolar cleft, is partly accessible to solvent. In contrast, Trp 227 is on a flexible loop near the active site, and the potential for extensive solvent quenching would explain its low fluorescence yield.
Identification of the Tryptophan Involved in Energy Transfer with NADPH-The emission spectra of the apoenzyme, the E⅐NADPH binary complexes, and the E⅐NADPH⅐testosterone ternary complexes of wild-type 3␣-HSD and each tryptophan mutant clearly identified Trp 86 as the tryptophan quenched by NADPH through an energy transfer mechanism (Fig. 3). The emission spectra of the binary complexes of the wild-type enzyme, the W148Y mutant, and the W227Y mutant each exhibited an energy transfer emission band of approximately the same intensity at 450 nm. The W86Y mutant lacked this band, indicating Trp 86 as the quenched tryptophan. This emission signal was unobservable in controls containing only 1.5 M NADPH at the same excitation wavelength (data not shown). The fluorescence spectra of the ternary complexes of wild-type enzyme and each tryptophan mutant showed additional quenching of protein fluorescence emission. With the exception of the W86Y mutant, there was a decrease in emission at 450 nm upon binding of testosterone to the E⅐NADPH complexes. Since the energy transfer band is assignable to Trp 86 , this decrease in emission indicates that steroid binding interferes with energy transfer between NADPH and Trp 86 .
Steady-state Kinetic Properties of the Tryptophan Mutants-The kinetic properties of the recombinant wild-type enzyme and the three tryptophan mutants were compared using four standard substrates: androsterone and NAD ϩ (oxidation reaction) and androstanedione and NADH (reduction reaction) (Table II). Mutation of Trp 86 altered both substrate and cofactor kinetics. The W86Y mutant resulted in a 3-fold increase in K m for androsterone and a 6-fold increase in K m for androstanedione. Likewise, this mutation increased the K m for NAD ϩ 6-fold and the K m for NADH 8-fold. In addition, of the three mutant enzymes, the W86Y mutant exhibited the lowest catalytic efficiency for all four substrates. The W227Y mutant displayed significantly impaired kinetics for each steroid substrate with 9-and 43-fold increases in K m for androsterone and androstanedione, respectively, but only slight differences with NADH (4-fold increase in K m ) and none with NAD ϩ . The inability to saturate the W227Y mutant due to solubility limitations of the steroid substrates emphasizes the importance of this residue in steroid binding and contributes to the low turnover numbers observed with this mutant. As expected from its position in the three-dimensional structure, the W148Y mutant gave only modest changes in kinetic constants compared with wild-type enzyme for all substrates.
Determination of the Binding Constants for NADPH and Testosterone-To directly evaluate the roles of each tryptophan in cofactor binding, we measured the K d values of each mutant for NADPH by fluorescence titration. For these experiments, NADPH is preferred over NADH, since its low K d value ensures saturation before interference from the inner filter effect occurs. Fig. 4A shows a typical titration curve of wild-type enzyme with NADPH and demonstrates saturation at the highest ligand concentrations. Titration of each tryptophan mutant indicated only minor changes in K d for cofactor (Fig. 4B) Equilibrium dialysis experiments quantitated the ability of each tryptophan mutant to bind radiolabeled testosterone to the E⅐NADH complex (Fig. 5). In these studies it was necessary to add NADH, since 3␣-HSD displays an ordered bi-bi mechanism in which cofactor binds first (28). NADH was used in these experiments because it is more stable than NADPH over the time required to reach binding equilibrium. Radiolabeled testosterone was chosen as the steroid for these experiments because it is bound by the enzyme, is not a substrate, and binding can be directly quantitated by scintillation counting. The binding constants for the wild-type enzyme and the W148Y mutant were similar, 4.2 Ϯ 0.8 and 6.6 Ϯ 0.9 M, respectively. The effects on steroid binding in the W86Y and W227Y mutants reflected the changes observed in the steady-state kinetic analysis. The W86Y mutant displayed a 7-fold increase in K d (31 Ϯ 8.6 M), and the W227Y mutant had a 22-fold increase in K d (92 Ϯ 32 M) for testosterone, indicating that both residues play an important role in steroid binding. Effect of the W86Y and W227Y Mutants on the Turnover of One-, Two-, and Three-ring Substrates-To elucidate the involvement of Trp 86 and Trp 227 in substrate recognition, the W86Y and W227Y mutants were used to turnover substrates of varying ring size: 4-nitrobenzaldehyde (one ring); 2-decalone (two rings); and 9,10-phenanthrenequinone (three rings) (Table II). The W227Y mutant was kinetically similar to wild-type enzyme, yielding similar K m values with 4-nitrobenzaldehyde and 2-decalone, and had a slightly elevated K m for 9,10phenanthrenequinone. However, mutation of Trp 86 altered the catalytic efficiency of the enzyme for each substrate, and the effects were related to the size of the molecule. Determination of the catalytic efficiency with 4-nitrobenzaldehyde was not possible for the W86Y mutant, since this was a very poor substrate with a k cat value 300-fold less than wild-type enzyme. This mutant gave a 65-and 14-fold reduction in catalytic efficiency for the turnover of 2-decalone and 9,10-phenanthrenequinone, respectively. As expected, the W148Y mutant had no effect on the turnover of one-, two-, and three-ring substrates.
Effect of W86Y and W227Y Mutations on the Affinity of 3␣-HSD Inhibitors-To further investigate the effects of Trp 86 and Trp 227 on ligand binding, we measured the ability of various 3␣-HSD inhibitors to block androsterone oxidation catalyzed by the W86Y and W227Y mutants (Fig. 6). Table III summarizes the IC 50 values obtained for various inhibitors. Values obtained for NSAIDs with recombinant wild-type enzyme were similar to those reported by Penning and Talalay for the native rat liver enzyme (15). Mutation of Trp 86 elevated the IC 50 values for all of the NSAIDs (indomethacin, sulindac, flufenamic acid, mefenamic acid, and meclofenamic acid) between 7-and 30-fold, while the IC 50 values of the same compounds with the W227Y mutant were not significantly altered. The W86Y and W227Y mutants displayed an 11-and 3-fold increase in IC 50 values for hexestrol, respectively. Mutation of Trp 227 affected inhibition by 1,7-and 1,10-phenanthroline, but inhibition by these compounds was unchanged in the W86Y mutant. Finally, neither tryptophan mutation significantly affected inhibition by the aldose reductase inhibitors zopolrestat and ponalrestat. These results suggest that NSAIDs and aldose reductase inhibitors bind to the same substrate binding pocket in a different manner. These findings have implications for structure-based drug design in the AKR superfamily. DISCUSSION We have described the first functional studies on amino acids located in the presumptive substrate binding site of rat liver 3␣-HSD. We used site-directed mutagenesis to probe the contributions of the three tryptophans in 3␣-HSD to protein fluorescence and the roles of each in cofactor, substrate, and inhibitor binding. Our results provide a model for steroid binding in 3␣-HSD and give insight into how different inhibitors may be accommodated in the apolar cleft.
Tryptophans, Protein Fluorescence, and Cofactor Binding-Quenching of intrinsic protein fluorescence upon NADPH binding has been observed in 3␣-HSD and used to calculate binding affinities for cofactor (18 -19). Mutation of each tryptophan produced only marginal changes in the K d for NADPH, and each mutant bound cofactor with nanomolar affinity. Since the cofactor binding site extends from the core of the barrel to the periphery of the structure and involves contacts with at least 12 amino acids, it is unlikely that these mutations caused gross structural changes in the three-dimensional structure of the protein. It was found that bound NADPH quenches the fluorescence of Trp 86 by an energy transfer mechanism. In the 3␣-HSD⅐NADP ϩ binary complex structure, Trp 86 is 10 Å away from the nicotinamide ring and is near enough for energy transfer to occur (29). Although removal of Trp 86 abolishes the energy transfer peak, some quenching of fluorescence emission is evident in the W86Y mutant and may result from collisional quenching of other residues in the cofactor binding pocket. A similar energy transfer mechanism occurs in aldose reductase, FIG. 5. Measurement of the binding of testosterone to the E⅐NADH complex of wild-type 3␣-HSD and W86Y, W148Y, and W227Y mutants. The binding of [ 3 H]testosterone to wild-type 3␣-HSD (A) and the W86Y (B), W148Y (C), and W227Y (D) mutants was measured by equilibrium dialysis in the presence of NADH. Experiments used 10 M protein with 2.5 mM NADH in 10 mM potassium phosphate (pH 7.0) buffer with 1 mM EDTA, 1 mM BME, and 4% acetonitrile as co-solvent. K d values were calculated from the binding data, as described under "Experimental Procedures." but the quenched tryptophan remains unidentified (30 -31).
The mechanism of fluorescence quenching in aldose reductase also involves a conformational change upon binding of cofactor (30). The structure of the aldose reductase apoenzyme showed that the tryptophan analogous to Trp 227 was packed against another loop of the protein, and that subsequent binding of NADPH disrupts this packing, thereby providing additional fluorescence quenching (32). In 3␣-HSD, it is not known if similar changes in the environment of Trp 227 occur upon cofactor binding, since the loop in which it resides was disordered in the apoenzyme structure and formed a crystal contact in the binary complex structure (16,17).
Tryptophans as Components of the Steroid Binding Pocket-Comparison of the fluorescence spectra of the binary and ternary complexes also provided evidence for the location of the steroid binding site. The emission spectra for the E⅐NADPH binary complexes of wild type and the W148Y and W227Y mutants gave a cofactor energy transfer emission band that was reduced in intensity upon binding of testosterone, indicating that bound steroid interferes with the energy transfer mechanism. Since the fluorescence of Trp 86 is quenched by the nicotinamide ring of NADPH, which transfers a hydride to the C3 position of the steroid, these data suggest that the steroid A-ring lies between this tryptophan and the cofactor.
The amino acids of the apolar cleft (Leu 54 , Trp 86 , Phe 128 , and Phe 129 ) would provide an ideal environment for interaction with steroid substrates. Previous studies with affinity labels and mechanism-based inactivators have targeted this general vicinity of the structure as the location of steroid binding by tagging a cysteine near but not in the proposed binding site (33)(34). The results of the kinetic studies with the W86Y mutant support the assertion that the steroid binding pocket of rat liver 3␣-HSD is the large apolar cleft near the catalytic tetrad (Asp 50 , Tyr 55 , Lys 84 , and His 117 ). The W86Y mutation affected the K m values for NAD ϩ and NADH and significantly altered the K m values for steroid substrates. In addition, our studies revealed that one-and two-ring substrates, which may occupy the same space as the A-and B-rings of the steroid, were turned over poorly by the W86Y mutant but that catalytic efficiency increased with substrate size. Turnover of 4-nitrobenzaldehyde was too low for accurate determination of a K m value, while the k cat /K m for 2-decalone was reduced 65-fold relative to wild type and was reduced even less for 9,10phenanthrenequinone and steroids. These data indicate that although the W86Y mutant may alter the topology of the apolar surface, larger substrates may compensate by interacting with a greater surface area of the cleft. The fluorescence data and kinetic studies on the W86Y mutant provide evidence that Trp 86 is near the A-and B-rings of bound steroid.
Unexpectedly, the K m values for androsterone and androstanedione and the K d for testosterone were increased dramatically in the W227Y mutant. Interestingly, this mutation did not affect the kinetic constants for smaller substrates. This size-specific effect implies that Trp 227 (and its associated loop) interacts with the C-and/or D-rings of steroid ligand.
Studies on aldose reductase suggest that the analogous tryptophan is part of the apolar cleft and is involved in substrate and inhibitor binding. In the three-dimensional structure of aldose reductase complexed with zopolrestat, the inhibitor occupies the apolar cleft, and this tryptophan makes van der Waals contacts with the ligand (35). Aldose reductase also catalyzes the reduction of the C21 aldehyde in isocorticosteroids (36). Modeling studies indicate that this tryptophan may interact with the A-and B-rings of the isocorticosteroid, which in this orientation would occupy the same space as the C-and D-rings of a 3-ketosteroid in 3␣-HSD (37). Also, in other (␣/␤) 8barrel proteins, loops on the C-terminal side of the barrel near the active site contribute to substrate binding (38 -40).
Our results would be consistent with the following model for steroid binding in 3␣-HSD (Fig. 7). The C3 position of bound steroid points toward the catalytic tetrad (Asp 50 , Tyr 55 , Lys 84 , and His 117 ) so that 4-pro-R-hydride transfer from the nicotinamide cofactor can occur. Orientation of the steroid ␣-face toward the side of the apolar cleft containing Trp 86 would preserve the known stereochemistry of hydride transfer. Also, a steroid in this position would interfere with energy transfer between Trp 86 and bound cofactor, as observed in spectra of the E⅐NADPH⅐testosterone complex. In addition, the length of the substrate would allow surface interactions between amino  acids along this side of the apolar cleft and the ␣-face of the steroid. Finally, Trp 227 (and its associated loop) could interact with either the edge or the ␤-face of the C-and/or D-rings of the steroid to form the opposite side of the binding cleft. Tryptophans analogous to Trp 86 and Trp 227 are present in other members of the AKR superfamily, including 17␤-HSD, 20␣-HSD, aldose reductase, and aldehyde reductase (8 -12). Based on the conservation of these residues in the superfamily and our site-directed mutagenesis data, these tryptophans are predicted to form basic components of the substrate binding pocket in the AKR proteins but probably do not determine specificity between steroid and sugar substrates.
Inhibitor Specificity: Discrimination between NSAIDs and Aldose Reductase Inhibitors-Inhibition of 3␣-HSD by NSAIDs is significant because of the high correlation that exists between the inhibition constants observed and the rank order of potency of these drugs (15). Since this rank order is essentially identical to that for prostaglandin H synthase (cyclooxygenase), the accepted target for NSAIDs, 3␣-HSD may also be a target for these agents. Importantly, NSAIDs are equipotent inhibitors of 3␣-HSD and prostaglandin H synthase. In contrast, aldose reductase inhibitors, e.g. zopolrestat and ponalrestat, are weak inhibitors of 3␣-HSD but potent inhibitors of aldose reductase. The ability of aldose reductase inhibitors to discriminate between 3␣-HSD and aldose reductase indicates that selective inhibition of AKR superfamily members is an achievable goal. Inhibition data obtained with the tryptophan mutants provides information on the nature of this specificity.
Structural elements of NSAIDs that are important in determining binding to 3␣-HSD and prostaglandin H synthase include 1) a carboxylic acid moiety (e.g. acetic acid or propionic acid) on an aromatic ring or indole, and 2) a methylene or carbonyl bridge to a lipophilic R-group (aromatic or aliphatic) (41). These requirements dictate the need for an anionic binding site located within a larger apolar pocket, and this is present in 3␣-HSD.
Penning and Talalay (15) pointed out the conformational similarities between indomethacin and androsterone. Based on this modeling, the carboxylic acid of indomethacin approximates the C3 ketone of a steroid and could fit into the anionic binding site of 3␣-HSD formed by Tyr 55 , Lys 84 , and His 117 . This site, which is conserved in the AKR superfamily, is part of the catalytic tetrad and is essential for polarization of the carbonyl group of the substrate. The p-chlorobenzoyl group of indomethacin was modeled perpendicular to the planar steroid and was superimposed on the C18 angular methyl group.
Our inhibition studies with the W86Y and W227Y mutants support aspects of this model. We have shown that Trp 86 is near the steroid A-ring and that the W86Y mutant decreases NSAID binding, indicating that the substrate carbonyl and the NSAID carboxylic acid may mimic one another and bind to the same site, as originally proposed. By contrast, the W227Y mutant dramatically affects steroid binding but does not alter inhibition by NSAIDs implying that the lipophilic R-group, exemplified by the p-chlorobenzoyl of indomethacin, may not bind in the same space as the C18 angular methyl group at the C/D-ring fusion.
The pharmacophor model proposed by Kador and Sharpless (42) for aldose reductase inhibitors is similar to that described for NSAIDs. Key features include 1) a primary aromatic region separated from an electrophilic group by 2.8 -3.8 Å; 2) a secondary lipophilic group containing one or two electronegative groups (hydroxyl or halogen); and 3) the ability to form a charge transfer complex with a polarized carbonyl, implying that an anionic binding site is required. These descriptors would predict that compounds like zopolrestat should inhibit 3␣-HSD and aldose reductase equally well, but zopolrestat inhibits aldose reductase with an IC 50 of 3 nM (43) and 3␣-HSD with an IC 50 of 46 M.
One explanation for this finding is that the apolar clefts of 3␣-HSD and aldose reductase are different. Zopolrestat binds to the aldose reductase⅐NADPH complex with 110 proteinligand contacts (35). In this complex, the carboxylate of zopolrestat occupies the anionic binding site formed by the active site residues, in accord with the model discussed above. But zopolrestat's binding avidity results from van der Waals interactions between four key residues (Trp 20 , Trp 111 , Phe 122 , and Leu 300 ) and the lipophilic moieties of the inhibitor. Over half of the protein-ligand contacts in this ternary complex are made with these four residues. The pthalazine ring is packed against Trp 20 , and a W20A mutant of aldose reductase dramatically impairs zopolrestat binding (31), while the benzothiazolyl ring is bound among Trp 111 , Phe 122 , and Leu 300 . In 3␣-HSD the anion binding site is conserved, but the residues that contact zopolrestat are substituted by Thr 24 , Phe 118 , Phe 129 , and Asn 306 , respectively. Structurally, these changes result in the elimination of one indole ring (Trp 20 ) and the substitution of another indole ring (Trp 111 ) by a phenyl ring. The ring bulk in the apolar cleft is thereby reduced, decreasing the number of protein-ligand interactions. As a consequence, the binding of zopolrestat is reduced to micromolar affinity.
The issue remains why NSAIDs and aldose reductase inhib- FIG. 7. Model of androstanedione binding in the substrate binding pocket of rat liver 3␣-HSD. This view displays the residues of the steroid hormone binding site in relation to the active site residues (Asp 50 , Tyr 55 , and Lys 84 ) and bound NADP ϩ cofactor (17). The location of Trp 86 on one side of the apolar cleft and the location of Trp 227 on a flexible loop can be seen. We have modeled androstanedione into the apolar cleft based on our mutagenesis results. It is proposed that the ␣-face of the steroid orients itself toward the side of the cleft containing Leu 54 , Trp 86 , Phe 128 , and Phe 129 and that the ␤-face of the steroid interacts with the loop containing Trp 227 . This figure was prepared using SETOR (44).
itors each bind to 3␣-HSD with micromolar affinity, but are affected differently by the tryptophan mutants. In the aldose reductase⅐NADPH⅐zopolrestat complex, tryptophans analogous to Trp 86 and Trp 227 contact the inhibitor (35). Mutation of Trp 86 in 3␣-HSD slightly increases the IC 50 value for both zopolrestat and ponalrestat but dramatically increases the IC 50 values for NSAIDs. Also, the W227Y mutant had no effect on inhibition by either the NSAIDs or aldose reductase inhibitors, suggesting that this residue does not interact with either type of inhibitor. Although the carboxylate moiety from both classes of inhibitors may bind at the anionic binding site formed by the active site residues, it is likely that the pthalazine ring of an aldose reductase inhibitor and the indole or benzyl ring of an NSAID that contain the carboxyl substituent fit into the apolar cleft differently.
We also used inhibitors that did not conform to either the NSAID or aldose reductase pharmacophor models. The planar 1,7-and 1,10-phenanthrolines, which lack any electronegative group, inhibit the wild type and tryptophan mutants extremely poorly, presumably because these compounds lack the substituents required for binding at the anionic site. In contrast, it is proposed that inhibitors, such as hexestrol, bind tightly because they are phenolic and contain the appropriate electronegative group for access to the anionic binding site.
These data suggest that the selective design of AKR inhibitors must not only take into account the need to accommodate the anionic binding site present in superfamily members, but in order to gain specificity it will be necessary to take advantage of differences in the apolar cleft. This would necessitate using multiring structures, like NSAIDs and the aldose reductase inhibitors, to ensure they occupy only the cleft of the targeted AKR protein.